| Full text | |
| Author(s): |
Gonzalez-Camacho, Lesly Alejandra
;
Kersul Faria, Joao Henrique
;
Machado, Lucas Torreao
;
Alves-Souza, Solange Nice
;
Rocha, A
;
Adeli, H
;
Dzemyda, G
;
Moreira, F
Total Authors: 8
|
| Document type: | Journal article |
| Source: | INFORMATION SYSTEMS AND TECHNOLOGIES, WORLDCIST 2022, VOL 3; v. 470, p. 19-pg., 2022-01-01. |
| Abstract | |
Recommender systems have gained importance and visibility mainly in e-commerce applications and the transmission of music and videos. In these platforms, the user can choose vast items, and recommender systems facilitate the users' choice by reducing the options to items of most significant interest. However, cold-start situations (new users in the system) make the recommendation difficult due to the lack of information about users' preferences. Social networking data can be used as information to reduce the cold-start impact. In this scenario, identifying the best and most influential friends can improve the recommendation by placing the group of friends with the most excellent affinity. Thus, using data from social networks as the primary source of external information to recommend items to cold-start users, a recommendation model was proposed based on the strength of friendship and the degree of influence between individuals. More specifically, with the new user's access into the system through the his/her social network credentials, we can identify his/her friends' groups and, among these, his/her most influential friends. The preference information of these significant users is used to recommend items (tracks) to the cold-start user. The proposal was validated using a controlled experiment in which 20 users effectively participated. A social network, built especially for the proposal, retained information about the interaction between friends on the social network and their access to a music streaming service. Users evaluated the recommender system, giving scores from 1 to 5 for each recommended song. The assertiveness of the model was computed using the Root Mean Squared Error (RMSE) metric, presenting a result of 1.57, which shows that the recommendation prediction was very close to the values given by users. The results also showed that the proposed model could be used to improve the recommendation of any user and not just cold-starts. Thus, the proposed model is quite adequate to improve the recommendation. (AU) | |
| FAPESP's process: | 14/04851-8 - Big Data and Business Intelligence - recommender system using social networks for cold-start issue |
| Grantee: | Solange Nice Alves de Souza |
| Support Opportunities: | Regular Research Grants |